The Race — And Power Struggles — Towards AGI
This week: Nvidia shares eye-popping 94% YoY revenue growth, Perplexity gets shopping.
Turns out even the brightest minds can't escape human nature. Emails from OpenAI's early days were just shared (thanks to the Musk v. Altman court case), and they reveal fascinating details about the last decade in the race towards AGI.
Hat tip: Habryka/LessWrong
The original vision of OpenAI sounded very noble — founded to be humanity's shield against an "AGI dictatorship." Back in 2015, the OpenAI founders were worried about Google and DeepMind hoarding all the power if they got to AGI first. As they said:
"AI should be an extension of individual human wills and, in the spirit of liberty, not concentrated in the hands of the few."
Plot twist: Building AGI is expensive. Like, crazy expensive. DeepMind was burning through $250M even back then in 2015. And here's where things get a little spicy — OpenAI, founded as a nonprofit champion of humanity, started eyeing that for-profit life. Why? To gather the capital and compute resources to rival Google in the AI race.
This is when things started to unravel with the OpenAI founders. You can read a curated thread at LessWrong.
Some observations/thoughts from me:
AGI might never be democratized. The compute horsepower and infrastructure needed means that this is limited to only a handful of companies on Earth who can afford it.
Open source AI — while principally most aligned with the original mission of OpenAI — seems tough to crack economically. While we have open source models from Meta and Alibaba that show signs of hitting some performance benchmarks, their distribution today pales in comparison.
Multiple sources at OpenAI, Anthropic, and Google hint that we're hitting a plateau with LLM improvements. Despite throwing more compute and data at the problem, the gains are slowing. It may be more like a natural pause in the invention cycles – similar to what biologists call punctuated equilibrium. But this may allow a new contender to pull ahead in the race.
This exchange makes me wonder: Are we doomed to be locked into the power struggle to shape humanity's future, or is there still hope for a more distributed future?
Thanks S.!
The Latest This Week
Nvidia reports record-breaking Q3 revenue as “age of AI” goes full steam: Nvidia posted $35.08 billion in revenue for Q3, a 94% year-over-year increase, with profits more than doubling to $19.31 billion. The results surpassed Wall Street expectations of $33.17 billion in revenue and 75 cents per share, with Nvidia delivering 81 cents per share. Despite these strong results, Nvidia’s stock dipped 1% in after-hours trading but remains up 195% year-to-date. Looking ahead, the company forecasts Q4 revenue of $37.5 billion, underscoring its dominance in the AI chip market.
Microsoft announced several new AI products at Ignite this week, from new Copilot Actions that automate daily tasks like summarizing meetings and compiling reports with simple prompts, to real-time language translation with Teams Interpreter. Azure AI Foundry is their AI “app store” ecosystem which provides developers with a robust toolchain for designing scalable AI applications. Additionally, they gave first looks at Windows 365 Link, a secure cloud-first device launching in 2025.
Perplexity, the AI search engine, introduced a shopping feature — seemingly going head to head with Google. On Monday, the company debuted a new shopping feature for its paid customers in the U.S. which offers shopping recommendations within Perplexity’s search results as well as the ability to place an order without going to a retailer’s website.
DeepSeek’s first reasoning model R1-Lite-Preview turns heads, beating OpenAI o1 performance: DeepSeek, an AI offshoot of Chinese quantitative hedge fund High-Flyer Capital Management focused on releasing high-performance open-source tech, has unveiled the R1-Lite-Preview, its latest reasoning-focused large language model (LLM). R1-Lite-Preview is said to offer performance nearing and, in some cases, exceeding OpenAI’s vaunted o1-preview model.
Mistral unleashes Pixtral Large and upgrades Le Chat into full-on ChatGPT competitor: Mistral has launched a slew of updates including a new, large foundational model named Pixtral Large. The company is further upgrading its free web-chased chatbot, Le Chat, adding image generation, web search, and an interactive “canvas,” matching the features of and turning it into a more serious and direct competitor to OpenAI’s ChatGPT.
ElevenLabs now offers ability to build conversational AI agents: ElevenLabs launched the ability to build conversational AI bots. The company announced that users can now build complete conversational agents on ElevenLabs’ developer platform, with customizable variables such as tone of voice and response length.
TikTok parent ByteDance reportedly values itself at $300B: ByteDance, the Chinese company that owns TikTok, valued itself at $300 billion in a recent share buyback offer, according to a new report in The Wall Street Journal. That number might sound familiar, as the WSJ previously reported on a ByteDance buyback offer at a $300 billion valuation back in September 2022. But a year later, the company’s valuation had reportedly fallen 26%, to $223 billion.